This Common Linux Practice is No Longer Best

Whenever IT architects, designers, and executives think about placement of new Linux applications on servers, the most common conclusion is to put the applications on x86-based servers. Frequently missing from the initial infrastructure decision-making process is a multi-year operational outlook that addresses growth and performance. Hence, what may be a best practice for low-growth business-critical systems or a low-usage, non-critical application may fall far short of being a best practice when the application grows to encompass tens, hundreds, or thousands of servers and is business-critical. Many IT professionals will claim that the scale-out x86 platform satisfies the performance and utilization requirements with the use of current virtualization platforms and strategies. However, this is not the case and the facts do not support it.

Furthermore, processor utilization is not always the gating factor for x86 servers; memory or channel utilization can impair performance. These bottlenecks and constraints greatly limit the number of applications and workload that can be placed on an x86 server, making it necessary to scale out to multiple processors and endure the ensuing increases in complexity and costs.

The new IBM Power Systems solutions built on POWER8 chips alters the landscape of possibilities by providing IT organizations with persuasive alternatives to the status quo. The IBM Power Systems singlesocket and two-socket servers are more efficient, cost-effective servers designed to run AIX and industry standard Linux workloads while delivering improved scalability, lower costs, and superior levels of service. IBM has aggressively priced these Power System offerings so that the TCA for comparable x86 one- and two-socket solutions can be 2x the cost of the Power System servers or more. Similarly, the TCO for the comparable x86 server solutions can be up to more than 300 percent more expensive.

There are numerous technical reasons to implement solutions on Power System servers beyond the financial ones. The POWER8 chip incorporates a Coherent Acceleration Processor Interface (CAPI). The CAPI port interfaces with the PCIe slot and enables external components such as accelerators and flash memory to directly communicate with the server’s memory. No operating system drivers are needed to utilize this feature. Initial IBM performance tests indicate a use case that shows a 24:1 CAPI Flash consolidation advantage over an x86 configuration designed to handle the same capacity while slashing costs by two-thirds.

Additionally, the Power System servers are optimized to support AIX and industry standard Linux operating systems (SUSE Linux Enterprise Server, Red Hat Enterprise Linux, and Ubuntu Server) by exploiting workload advantages of POWER8, PowerKVM (Linux only) and PowerVM – yielding better performance, RAS (reliability, availability, and serviceability), and virtualization by being tuned to the environment. IT executives can order these solutions factory pre-loaded and get them delivered with an installation toolkit so that they can quickly become operational.

The PowerVM hypervisor offers capabilities beyond those obtainable on a commodity server with features such as active memory sharing, dynamic logical partitioning, and live partition mobility. With live partition mobility applications can be moved with minimal effort and without application shut down. This reduces the number of outages, allows for quiescing processors during low levels of utilization to save on power consumption, or swapping of servers without impact to business operations. Moreover, unlike VMware’s vSphere5.1 which only allows 96 GB of memory per socket, PowerVM does not have a virtual memory cap. Memory limitations can be problematic for a number of workload types that require a lot of memory. With PowerVM POWER8 servers can increase workloads within a single system up to 32 times to achieve superior memory and processor utilization levels.

The new IBM PowerKVM hypervisor provides an open virtualization platform choice for scale-out Linux users on the POWER8 servers. It is managed like any other KVM host and comes with open source tools so that users can avoid proprietary vendor lock-in. PowerKVM is designed to exploit POWER8 features like “micro-threading” and symmetric multithreading, with up to eight threads per core. The hypervisor optimizes real memory and allows for memory to be overcommitted and swapped out when inactive.

Additionally, Power Systems’ increased utilization and scaling capabilities reduce data center sprawl and offer multiple alternatives to maintaining, and even reducing, footprint requirements at refresh cycles. Furthermore, they will also reduce data replication tasks, enable systems to handle workload spikes, and simplify disaster recovery; and these characteristics end up shrinking the synchronization time needed in the nightly batch window. The net is all these Power Systems features streamline operations and costs while improving service to users thereby enabling administrators to focus more on revenue-generating and performance-enhancing tasks instead of ongoing operations. It will be years before x86 solutions can match today’s Power System capabilities.

The above was an excerpt from: The IBM Power Scale-out Advantage

Flagship Solutions Group has the expertise to answer any questions you might have relating to Linux and the benefits that can be achieved with IBM Power Systems. Schedule a consultation today for assistance.

If you liked this blog, you might also like: Get ready. Get set. Linux on POWER is now a real option… Go!

Stay connected online: